Apple's Writing Tools (proofread, rewrite, tone) are a system feature: select text, open the menu, pick an operation, review the result. OpenCues covers similar ground with a different philosophy: the request lives inside the text itself, on any platform, with any model.
The key differences
- Invocation. Writing Tools are menu-driven: select, summon, choose. OpenCues is typed:
make this formal _where your fingers already are, plus continuous modes (agent tasks) that a menu can't express. - Platform. Writing Tools exist on Apple Intelligence-capable Apple devices, in most apps with standard text fields. OpenCues runs in its supported hosts, on macOS and Linux (WSL2 on Windows).
- Model choice. Writing Tools use Apple's models, plus an optional ChatGPT compose hookup; there's no bring-your-own-model. OpenCues is model agnostic: Cerebras, Groq, OpenAI, Anthropic, Gemini, OpenRouter, or fully local via Ollama.
- Granularity. Rewrite swaps your selection for a new version (with a revert to original); Proofread underlines changes for one-by-one review. OpenCues edits land per-word, dimmed, each one revertable by cycling; your original text is always one cycle away.
- Openness. One is an OS feature; the other is Apache-2.0 code plus an open standard anyone can implement.
Where Writing Tools shine is zero setup on a recent iPhone. Where OpenCues shines is everywhere else: terminals, AI prompt boxes, your own model, your own extensions.
Two more axes: privacy and reach. Writing Tools process on device; OpenCues gets a similar guarantee on cloud calls by tokenising identity values before dispatch, or goes fully local via Ollama. And Apple covers iPhone and iPad, which OpenCues does not: there is no mobile host today.
Where Writing Tools struggles
The selection-popover flow is five or six discrete steps, and it only edits what already exists: there is no generation while typing and no way to summon data. Revert exists only while the popover session is open; after that it is the app's undo. Apple silicon is the ticket in, Apple's models are the only models, and the whole feature set is fixed.
By the numbers
The measured flow is the longest of the native tools: six steps to compose (Control-click, menu, Compose, describe, Return, Done) and five to rewrite a selection, against one step for either with OpenCues. Revert exists only inside the popover session; after Done it is the app's undo. Counts and sequences in the full comparison.
Related: OpenCues vs Grammarly · Fix grammar as you type · OpenCues vs Raycast AI · GitHub repo